System for processing a quantized vector

Description

[0001] The present invention relates to a decoder for quantized-vector using vector quantization and capable of compressing information sources which strongly correlate to one another, such as audio information or image information. This specification also discloses a method of making the code book therefor, a system for transferring quantized vector, and an image processing apparatus utilizing the vector quantization.

[0002] Vector quantization has been known as an efficient non-permanent information type encoding method.

[0003] Vector quantization which handles a plurality of information data always confronts a critical problem of reducing the amount of input data since the processing time and the size of the hardware respectively increase exponentially in proportion to the number of data items to be handled.

[0004] For example, image data is sometimes vector-quantized in accordance with the following steps: cutting image data in the block units; rendering the image data to be subjected to orthogonal transform such as discrete Fourier transform and discrete cosine transform so as to convert the image data into frequency data; and thus vector-quantizing the converted data. The above-described vector-quantization utilizes the characteristics of the orthgonally transformed image data whereby energy is concentrated in the low frequency band and thereby redundancy can be restricted. When image data is vector-quantized, the high frequency portion from which intense energy rarely outputs is subjected to a masking processing in addition to the above-described orthogonal transform.

[0005] As a result of the above-described processes, the number of data items is reduced during the first stage of the overall vector quantization. Proposals concerning the above-described vector-quantization have been made in USSN 261,276 and 312,775. (Now issued as US 5,157,743 and US 5,162,923 respectively.)

[0006] If the above-described method of masking the high frequency portion of the image is employed, the amount of input data for vector quantization is reduced. Therefore, the quantity of calculations required to retrieve the output vector can be restricted. Also during the above-described retrieval process, the vector data in the code book, to which reference is made, usually has no high frequency data similarly to the input data. The reason for this lies in that the expected value of the conversion coefficient of the base, in which a high frequency is present, is "0" when the overall image to be input is considered.

[0007] The above-described vector quantization in which the high frequency data is removed will cause a relatively high quality quantized image to be obtained even if it is applied to a usual image. However, critical quantization errors cannot be prevented in the portions having sharp edges or thin line structures. Furthermore, some images such as fine images of one pixel width cannot be reproduced at all without the high frequency components.

[0008] The deterioration in image quality due to the above-described quantization errors will become even more critical problem due to the wide use of highly precise images.

[0009] It might be considered to employ processing of supplying a noise component, whose average level is zero, to the high frequency region as a method of compensating the masked high frequency component. However, the above-described method cannot substantially overcome the above-described problem, and, what is worse, a further critical error can occur. Although the above-described problem can, of course, be overcome if the structure is arranged in which the above-described masking of the high frequency band is not employed, the vector quantization cannot be put into practical use since the processing system increases exponentially in proportion to the number of the frequency components to which a reference is made.

[0010] An object of the invention is to provide a quantized vector decoding apparatus capable of generating an output vector approximating the original information when the output vector is generated from a vector-quantized code.

[0011] An alternative object of the invention is to provide a system for transferring a quantized vector capable of both raising the retrieval speed in a code book and improving the quality of the output vector on the decoding side when the quantized data is generated or encoded.

[0012] A further alternative object of the invention is to provide a signal processing apparatus capable of improving the quality of the data in the high frequency band of the input signal.

[0013] A still further alternative object of the invention is to provide a vector quantizer capable of changing the code book, for use in the vector quantization processing, in accordance with the result of a classification of the characteristics of the high frequency information of the input data.

[0014] The invention provides a decoding apparatus comprising: a decoding apparatus comprising means for decoding a code which encodes a first set of frequency components of a signal, characterised in that said decoding means adds a second set of frequency components which was not encoded in said received code to the first set of frequency components using an assumed correlation between the first and second sets of frequency components. The invention may be applied for example to audio or image signals.

[0015] Embodiments of this invention exploit the fact that a plurality of signal values arranged to be adjacent to one another in a one-dimensional direction or a two-dimensional direction so as to form a block tend to be correlated to one another.

[0016] According to embodiments of the present invention, only the first set of frequency components are required at the time of the retrieval. Therefore, the retrieving speed can be raised and the two components portions can be transmitted. As a result, the quality of reproduction of the original signal can be improved.

[0017] In some embodiments of the present invention, the Hadamard transform is used as an orthogonal transform.

[0018] According to such embodiments of the present invention, the output means includes means for subjecting the output vector to inverse orthogonal transform.

[0019] In embodiments of the invention, code books of the encoding side device and the decoding side device can be made a common type in terms of the first output vector component portion (first set of frequency components). Therefore, the decoding operation can be reliably performed. Furthermore, since the second output vector component portion (second set) is omitted, the retrieving speed in the encoding side can be raised or the capacity of the code book storage means can be reduced.

[0020] Other objects and advantage besides those discussed above shall be apparent to those skilled in the art from the description of a preferred embodiment of the invention which follows: In the description, reference is made to accompanying drawings, which form a part hereof, and which illustrate an example of the invention. Such example, however, is not exhaustive of the various embodiments of the invention, and therefore reference is made to the claims which follow the description for determining the scope of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

[0021] 

Fig. 1 is a block diagram which illustrates a first embodiment in which the present invention is applied to an image signal encoder/decoder;

Fig. 2A illustrates the structure of the matrix elements which can be obtained by Hadamard transformation;

Fig. 2B illustrates the structure of the sequency data which can be obtained by Hadamard transformation;

Fig. 2C illustrates a mask for masking the high frequency component and that for masking the DC component;

Fig. 3 is a block diagram of a vector quantizer for the encoding side according to the first embodiment;

Fig. 4 illustrates the bit distribution in a scalar quantizer 31 included in a vector quantizer 4 according to the first embodiment;

Fig. 5 illustrates the structure of a code book 5b for the decoding side according to the first embodiment;

Fig. 6 illustrates the structure of an image quality improving apparatus exploiting the principle of the present invention;

Fig. 7 is a block diagram which illustrates an embodiment in which the present invention is applied to a copying machine;

Fig. 8 is a perspective view which illustrates the appearance of the copying machine shown in Fig. 7;

Fig. 9 is an internal structural view of the copying machine shown in Figs. 7 and 8;

Fig. 10A is a block diagram which illustrates the structure of a compressor 303;

Fig. 10B is a block diagram which illustrates the structure of an L* compressor;

Figs. 11A to 11D illustrate the masking position in a high frequency masking circuit;

Fig. 12 is a block diagram which illustrates the structure of a vector quantizer 407;

Figs. 13A to 13D are tables illustrates the structure of data in a code book 408 according to the enbodiment of Figs. 7-12:

Fig. 14 is a block diagram which illustrates the structure of an expander according to the embodiment of Figs. 7-12; and

Fig. 15 illustrates the base function of the Hadamard transformation for use in embodiments of the present invention.

[0022] In the following description, the term "sequency" will be used to refer to spatial frequency data.

[0023] According to the present invention, information suitable for collectively vector-quantizing a plurality of data items correlating to one another is handled, the information being image information or audio information. Then, embodiments in which the present invention is applied to vector-quantization of image data will be described.

<Description of Principle>

[0024] In general, image data items strongly correlate to one another. Spectrums, generated when an image is cut into uniform blocks and the inside portion of each of the blocks is frequency-analyzed, show patterns which can frequently appear. Therefore, there is a high frequency spectrum which can be easily generated with respect to a certain low frequency spectrum. That is, there is the relationship between the high frequency portion and the low frequency portion of a spectrum obtained by frequency-analyzing one image block.

[0025] Image data specified by p which has been cut in block units is usually expressed by BLKp obtainable from:


   where n: number of data items in a block
      fi: base of orthogonal transform
In this state, it is assumed that the more i of fi, the more its frequency is. The spectrum B_p of BLK_p is expressed by:

[0026] The expected value of a_pj with respect to overall vectors (p = 1, 2, 3,...) of the input image is zero when conversion coefficient a_pj of the j-th (j >> 1) high frequency is considered. However, when only an input image whose low frequency band spectrum shows a certain pattern, the conversion coefficient apj of the high frequency must be distributed in a region centered at a certain value due to the correlation with a low frequency spectrum. Furthermore, when energy of the low frequency band is large, also energy of the high frequency band is large. Therefore, the expected value of apj which appears together with the lower frequency band pattern with large energy and obtained by changing p is not zero.

[0027] It is assumed that the spectrum for a block (specified by number p) of an image to be input is S_p (dimension n), while a spectrum S_p1 (dimension m) relates to low frequencies and a balance spectrum S_pn (dimension n-m) relates to a high frequency. Furthermore, it is assumed that the conversion coefficient a_pj (j >> 1) belongs to S_pn.

[0028] That is, the following relationships are held:

[0029] As is shown, the average value of the j-th coefficient a_pj obtained by changing p for the overall input image is zero. On the other hand, it is assumed that dispersion of the j-th coefficient a_pj for the overall blocks of the input image is V_j. Furthermore, it is assumed that a certain quantized vector corresponding to input image block spectrum S_p is:

(1) First, error when mapping from S_p to C_p is performed is considered. The mapping from S_p to C_p corresponds to the conventional technology in which the high frequency band is masked (cut). The square error expected value of a_pj of all of the inputs, over all values of p, becomes: V_j

(2) Then, errors, when it is assumed that N_p is a set of high frequency noise vectors and mapping from S_p to C_p + N_p is performed, are considered. The above-described case corresponds to the case in which high frequency noise is added at the time of the decoding process described in the prior art. That is, assuming that a set of high frequency noise vector N_p is expressed by:

and dispersion of coefficient n_pj of the high frequency of this set for overall inputs is V_j (which is the same value at the dispersion of a_pj), the square error expected value of a_pj of all of the inputs becomes: 2V_j
Therefore, if high frequency noise is added, the level of errors becomes enlarged in comparison to a case in which the high frequency band is cut.

(3) Then, a case where a mapping from S_p to C_p + E is performed is considered. In this consideration, E is defined to be as follows. Image vector set consisting of input image data of which low frequency portion is mapped to quantized vector C_p is called Q_p expressed by:

that is, Q_p ε S_p and mapping from Q_p to C (masking) is performed. Assuming that the average and the dispersion of q_i for all of the inputs (all values of p) are e_i and v_i, respectively, and E is defined as follows:

Assuming that a mapping from S_p to C_p + E has been performed in the case of a certain input S, the square error expected value of a_pj becomes v_j.

(4) Then, the case in which a mapping from S_p to C_p + Q_pn is considered. It is provided that only the high frequency portion of the image vector set Q_p, formed by mapping the above-described low frequency portion to C_p, and expressed by:

is subjected to the mapping from S_p to C_p + Q_pn. The square error expected value of a_pj becomes 2v_j if Q_pn belongs to Q_p regardless of Q_pn.

[0030] It is expected for v_j to be smaller than V_j. Therefore, the quantization error of the mapping from S_p to C_p + E can be made smaller than the mapping from S_p to C_p. Furthermore, v_i becomes smaller if the correlation between the spectrum vector S_pi in the low frequency region of the input image and the high frequency conversion coefficient a_pj is strong. As described above, error when the mapping from S_p to C_p is performed is 2V_j. On the other hand, error when the mapping from S_p to C_p + E is performed is v_j, and error when the mapping from S_p to C_p + Q_pn is performed is 2v_j. Therefore, if there is a strong correlation which meets the following relationship:


quantization error will be less than when mapping from S to C_p is performed, if mapping is performed with an optional Q_pn as follows:

[0031] Thus, the image quality can be improved by applying the present invention to the vector quantization of the image. Then, a system for vector-quantizing an image, transferring the vector-quantized image code, and decoding the received code (a first embodiment), an apparatus in which above principle is applied to an improvement in image quality, and an apparatus according to an embodiment in which the present invention is applied to a copying machine will be described.

First Example: Application to an Image vector-quantizing/decoding apparatus

Structure of the apparatus

[0032] Now, an embodiment in which the present invention is applied to vector quantization of an image signal will be described with reference to the drawings.

[0033] Fig. 1 is a block diagram which illustrates the overall structure of an image compressing/decoding system according to the first embodiment of the present invention. The image compressing/decoding system comprises a transmission side including an encoder 50 and a transmission channel 23, and a receive side including a decoder 60. Image data input through an image data line 20 is aligned into block data each of which is formed by 4 pixels x 4 pixels by a line buffer 1 before subjected to orthogonal conversion by an Hadamard converter 2. As a result of this orthogonal transformation, the block data of 4 pixels x 4 pixels is transformed to a DC component data DC and 15-dimensional AC component data AC (A₁ to A₁₅) as shown in Fig. 2A.

[0034] The DC component data DC is supplied to a scalar quantizer 3 in which it is scalar-quantized so as to be converted into the code of the scalar quantization level. The code of the DC component is supplied to a transmitter 6 via a signal line 21. On the other hand, the AC component data AC is vector-quantized in a vector quantizer 4. That is, the AC component data AC is subjected to the vector-quantization due to the reference made to the low frequency data of the AC component data AC, the reference being made in the vector quantizer 4. As a result, the vector code can be generated, the vector code being then supplied to the transmitter 6 through a signal line 22.

[0035] The transmitter 6 collects the above-described codes to the decoder 60 via the transmission channel 23.

[0036] The receiver 7 divides the data thus transmitted into a scalar-quantized level code, which is a DC component of the image vector, and a vector code which is an AC component so that the scalar-quantized level code is supplied to the decoder 8 and the vector code is supplied to the decoder 9 via corresponding lines 24 and 25. The scalar decoder 8 generates the actual DC component value in accordance with the received code so as to output it to the Hadamard converter 2. The vector decoder 9 generates low frequency vector data and high frequency vector data in accordance with the received AC component code of the image so as to output them to the Hadamard converter 2. In this state, it must be considered that the vector code decoder 9 generates not only the low frequency vector code but also the high frequency vector code which was not used at the encoding operation in the encoder 50.

[0037] A Hadamard inverse converter 2 subjects the received actual DC component value and the low frequency and the high frequency vector data to inverse Hadamard transformation so that block data of an actual image is formed again. The block data which has been formed again is serially supplied to a signal line 26 via a parallel to serial converter 10.

[0038] As described above, the structure and the operation of the apparatus shown in Fig. 1 are arranged. Then, the operation of the encoder 50 and the decoder 60 will be described in detail.

Vector-quantization on the encoding side

[0039] Sixteen items of sequency data resulting from the Hadamard transformation are aligned into block data shown in Fig. 2A depending upon the level of the frequency (see also Fig. 23). An element (1, 1) of a block formed by 4 x 4 sequency components the DC component which is subjected to the scalar quantization by the scalar quantizer 4 as described above.

[0040] The remaining 15 sequency data items (A₁ to A₁₅) are considered as a 15-dimensional vector Therefore, they are subjected to the vector quantization by the vector quantizer 4 in accordance with a code book stored in a ROM 5a. Basically, a code book 5a for the vector quantization provided for the encoding side and a code book 5b for decoding provided for the decoding side are the same code books. These code books 5a and 5b are representative 15-dimensional vectors which have been previously obtained by "training" using a multiplicity of images.

[0041] The high frequency portion of the 15-dimensional AC component data to be supplied to the vector quantizer 4 are uniformly masked as shown in Fig. 2C. According to this embodiment, the high frequency portion is, for example, determined to be A₃, A₆, A₇, A₉ to A₁₅ as shown in Fig. 2B. Therefore, A₁, A₂, A₄, A₅ and A₈ are considered to be the low frequency portion. Thus, the AC component of the image data are reduced from the 15-dimensional data to 5-dimensional data after the orthogonal transformation.

[0042] The vector quantizer 4 selects a vector which is the nearest to the input vector of all of the vectors registered in the code book 5 so as to supply its index code to the line 22.

[0043] According to this embodiment, the nearest vector is selected by employing a Look Up Table method with a ROM in which the scalar quantization is also used as shown in Fig. 3 as an alternative to distance calculations.

[0044] Fig. 3 is a block diagram which illustrates a structure in which the vector quantizer 4 and the ROM 5a shown in Fig. 1 are arranged in accordance with the Look Up Table method. Referring to this drawing, the structrure comprises a high frequency masking circuit 30, a scalar quantizer 31 and a ROM 5a. Referring to Fig. 3, the high frequency portion of the AC data which has been Hadamard transformed is uniformly masked to 0 in the high frequency masking portion 30 as shown in Fig. 2C. Data of elements A1, A2, A4, A5 and A8 which have not been masked are supplied to the scalar quantizer 31 in which their lower bits are cut in such a manner that a bit distribution as shown in Fig. 4 can be realized. That is, the number of bits of each of the elements A1, A2, A4, A5 and A8 is arranged to be "4", "3", "4", "3", and "3", respectively, and the total number of bits becomes 17. An output vector (the nearest vector) for all of the input states of the 17 bits has been previously carried out, and its index value has been written in the ROM 5a. Thus, the index of the nearest vector of the input vector can be immediately transmitted in the vector quantizer 4.

[0045] As described above, the calculation speed in the scalar quantizer 31 can be raised by applying a mask to the high frequency region before the quantization in the vector quantizer 4 as shown in Fig. 3. Furthermore, the size of the capacity of the ROM 5a can be reduced since it must store only low frequency data.

Vector reconstruction on the decoding side

[0046] The reproduction of the output vector in the decoding side vector decoder 9 will be described with reference to Fig. 5. The decoder 9 receives only the index code for vector-quantization obtained by the encoding side vector quantizer 4 in accordance with the low frequency component of the image data. The decoder reads out output vector data from the table memory 5b in accordance with the index code thus received. The output vector stored in the ROM 5b includes not only the vector data for low frequencies but also the vector data for high frequencies as shown in Fig. 5. The data read out at this time includes high frequency data which is expressed with the mask shown in Fig. 4. That is, since the ROM 5b reserves in its output vector not only the low frequency component but also the high frequency component, the reproduced image cannot deteriorate.

Second Example: Application for Improving the Image Quality

[0047] Fig. 6 illustrates the concept of an image quality improving apparatus in which the principles described above are applied to obtain an improvement in image quality. That is, the system shown in Fig. 1 (the first embodiment) acts to compress the input image by vector-quantization, and decoding and reproducing the thus compressed information after transferring or storing it. On the other hand, the image quality improving apparatus according to the present example acts to improve image quality by modifying the image in such a manner that the reliability of the high frequency component of an image is improved, the image being an image lacking reliability in the high frequency data thereof,such as an image lacking the high frequency data, an image from which the high frequency data is omitted or an image having a high frequency region mixed with noise.

[0048] Referring to Fig. 6, image data obtained by cutting the image in block unit of m by n pixels is supplied from a line 40. The image data cut in block units is orthogonally converted by an orthogonal converter 41. The orthogonal converter 41 may be the same as the Hadamard converter 2. If the converter 41 is the same as the converter 2, the sequency data has the same structure as that of that shown in Fig. 2A and 2B. The sequency data is supplied to a high frequency masking portion 42. The determination of the sequency data to be masked is made by data stored in a masking position memory 43 after selecting it from the 15 sequency data items. The high frequency data whose reliability is insufficient is cut by the high frequency masking portion 11. The masking position memory is constituted of storage means such as a ROM. The masking position memory 43 stores, for example, the position of the element to be masked as shown in Fig. 2C. In the case of the Hadamard converter, since the elements disposed in the lower right portion are in the higher frequency band, data stored in the masking position memory 43 acts to determine the reliable frequency level of the spectrum data. When AC components of the output from the orthogonal converter 41 are expressed by L + H since they are sum of sets of the high frequency components and the low frequency components, the output from the high frequency masking portion 42 becomes L.

[0049] The data L transmitted from the high frequency masking portion 42 is supplied to a retrieval portion 45 for retrieving the nearest vector, and in parallel is also supplied to a synthesizing portion 48. The synthesizing portion 48 synthesizes the output from the high frequency masking portion 42 and that from the low frequency masking portion 46. In the retrieval portion 45, the distances from all of the vectors registered in the table memory 5b to the data L from the masking portion 42 are calculated. As a result, the vector which is positioned at the shortest spatial distance is transmitted to the low frequency masking portion 46. That is, the vector retrieval performed in the retrieval portion 45 is a similar operation to pattern matching rather than vector quantization. The nearest vector to the vector formed by only the low frequency sequency which is considered to be reliable is retrieve a into which high frequency sequency has been added. This is supplied to the low frequency masking portion 46, where the vector which has been retrieved in the retrieval portion 46 as the nearest to the input vector L is expressed by L′ + H′.

[0050] On the other hand, the low frequency masking portion 46 masks the low frequency band of the vector data (L′ + H′) produced by the vector retrieval. That is, the output from the masking portion 46 becomes H′. The output L from the high frequency masking portion 42 and the output H from the low frequency masking portion 46 are supplied to the synthesizing portion 48 in which they are synthesized. That is, the output from the synthesizing portion 48 becomes as follows:


In other words, the high frequency data in the sequency data is replaced by the high frequency masking portion 42, the retrieval portion 45 and the low frequency masking portion 46. The thus replaced data is returned to an actual image data by the orthogonal converter 47 so as to be transmitted.

[0051] Since H′ thus replaced is the high frequency portion of the vector which has been retrieved as a vector which has a strong correlation with the reliable L, L + H′ becomes improved in its high frequency band.

Further example: Application to a Copying Machine

[0052] Fig. 7 is a block diagram which illustrates a third embodiment of an apparatus structured in such a manner that the present invention is applied to a copying machine.

Schematic structure of the copying machine

[0053] The copying machine comprises a reader section 100, a memory section 300 and a printer section 200.

[0054] As described above, reference numeral 100 represents the reader section, 115 represents a read sensor and 118 represents a light source for illuminating an original. Reference numeral 106 represents an amplifier portion for amplifying an analog image signal transmitted from the read sensor 115. The amplified analog image signal is supplied to an A/D converter 107. In this structure, the read sensor 115 is a CCD color sensor to which pixels each of which is represented by R (Red), G (Green) and B (Blue) signals are serially transferred, and they are transmitted therefrom. Reference numeral 108 represents a latch circuit for latching and holding the B, G and R signals in this sequential order to simultaneously transmit the color signals B, G and R for one pixel. Reference numeral 109 represents a color conversion portion in which L*, a* and b* signals are created from the B, G and R signals, the color conversion portion 109 being structured in the form of a look up table made by, for example, a ROM (Read Only Memory). Therefore, the output signal from the reader section 100 becomes L*, a* and b* signals.

[0055] Reference numeral 300 represents a memory section. The memory section 300 comprises a compression circuit 303, an expansion circuit 304 and an interface circuit 320 in addition to an image memory 301. The L* (brightness signal), a* (color signal) and b* (color signal) are independently compressed in the compression circuit 303. Finally, information about L*, a* and b* about a predetermined number of pixels is compressed to 32 bits so as to be supplied to the image memory 301.

[0056] The detailed operation of this will be described later.

[0057] The image signal stored in the image memory 301 is transmitted through a communication interface 320 or transmitted to the printer section 200. In the case where it is transmitted to the printer portion 200, the compressed image signal stored in the image memory 301 is read out by an expander 304 and is decoded so as to return to the original L*, a* and b* signals before transmitted to the printer portion 200.

[0058] The printer portion 200 receives the L*, a* and b* signals transmitted from the memory portion 300 in a color conversion portion 255 thereof. The color conversion portion 255 supplies the L*, a* and b* signals with switching them to laser units (201C, 201M, 201Y and 201K) in response to a command of non-synchronized write-timing signal (omitted from illustration) issued from a plurality of photosensitive bodies (211C, 211M, 211Y and 211K). Also, the color conversion portion also converts the image signals in the form of the L*, a* and b* signals into image signals representing toner C (cyan), M (magenta), Y (yellow) and K (black).

[0059] The laser unit according to this embodiment comprises 4 portions, 201C, 201M, 201Y and 201K so that image of C (cyan), M (magenta), Y (yellow) and K (black) is written in the photosensitive bodies (211C, 211M, 211Y and 211K). The bodies 211C, 211M, 211Y and 211K are each in the form of a drum and disposed so as to correspond to the respective laser unit (201C, 201M, 201Y and 201K).

[0060] Numeral 1100 denotes a operator section. CPU 1150 connected to the section controls the reader, memory and printer sections in accordance with operator's instructions. For example, the contents of image memory 301 are erased in accordance with a memory clear signal 1155. Several signals input into the CPU 1150, such as power switch signal 1152 for the printer section, APC ready signal from a APC circuit 1004a which controls the laser power, and registration ready signal 1158 from an automatic registration correction circuit 1401.

[0061] Fig. 8 is a perspective view which illustrates the appearance of the digital color copying machine shown in Fig. 7. Referring to Fig. 8, as described above, reference numerals 100 represents the color reader section for storing an image signal, 300 represents a memory section for storing a compressed image signal and 200 represents a color printer section for forming a color image and transmit it. The copying machine according to this embodiment comprises a sorter 1300, an A·D·F (Auto Document Feeder) 1200 and a paper deck 1401 as option equipments.

[0062] The sorter 1300 aligns color copying paper sheets. The an A·D·F 1200 automatically supplies the original sheet. The deck 1401 stocks and supplies a large quantity of copying paper. Reference numerals 1400A and 1400B represent paper feeders which are, according to the present invention, in the form of vertical double cassettes.

[0063] Fig. 9 illustrates the internal structure of the copying machine according to the third embodiment. The operation of each of the portions will be described later. Since Fig. 9 shows the minimum structure required to operate basically, the option equipments such as the sorter 1300 or the paper deck 1401 shown in Fig.8 are omitted from the illustration.

[0064] Then, the reader section 100 will be described in detail with reference to Fig. 9. Referring to Fig. 9, reference numeral 112 represents an original illumination lamp serving as a light source, 113 represents a reflecting mirror, 114 represents a rod lens array for converging light reflected from the original 122 placed on the original setting glass 121 on to the read sensor 115. A scanning body 118 includes the original illuminating lamp 112, the reflecting mirror 113, the rod lens array 114, the read sensor 115 and an A/D conversion circuit 117. The A/D conversion circuit 117 A/D-converts the image signal transmitted from the read sensor 115 through a signal line. The above-described components 112 to 117 included by the scanning body 118 and the scanning body 118 perform linear actions with integrated to one another in the direction designated by an arrow A. A control circuit 120 stores, calculates and transmits an image signal of a predetermined number of bits, which is 8 bits according to this embodiment, transmitted from the A/D conversion circuit substrate 117 through a signal line 119.

[0065] As shown in Fig. 9, the color printer 200 in the form of a 4-serial drum type is structured by four units (205C, 205M, 205Y and 205K) centered at the photosensitive drums (211C, 211M, 211Y and 211K). Each of the units has a color developing machine each of which includes toner of independent colors. Referring to Fig. 9, the laser oscillators (201K, 201Y, 201M and 201C) correspond to color signals respectively representing K (black), Y (yellow), M (magenta) and C (cyan).

[0066] The copying machine includes a polygon mirror (omitted from illustration) so that laser beams scan in the main scanning direction when the polygon mirror rotates. Thus, one line is formed.

[0067] The photosensitive drum 211 for 1 unit rotates clockwise so that the surface of the photosensitive body thereof is charged by a charger 212. Thus, image information is written on the drum due to the switching of/off of the laser beam. Although the image thus formed is a latent image, the latent image is made visible (an apparent image) when color toner is adhered thereto by a development sleeve 206. The thus formed apparent image is transcribed by a transfer charger 210 to a copying paper supplied by a supply roller 207 from the paper cassette 208. In this case, the margin of the image is adjusted by a register roller 250.

[0068] The copying paper is conveyed by a conveyance belt 209 from the development unit 211C for C (cyan) to the development unit 211M for the next color M (magenta). Thus, the toner images are overlapped successively during the conveyance through the development unit 211Y for Y (yellow) and the development unit 211K for K (black). When the color toner of the four color has been adhered on the copying paper, the copying paper is transferred to a heating fixing device 213 at which the color image is fixed by heat. The copying paper is then placed on a copy tray 215 by an paper-exhausting roller 216.

Compression portion and Expansion portion

[0069] 

[0070] The compressing portion 303 and the expansion portion 304 shown in Fig. 7 will be described in detail.

[0071] The compression portion 303 shown in Fig. 7 is structured as shown in Fig. 10A so that the compression L* and compressions a* and b* are independently performed. Fig. 10B illustrates the internal blocks of the L* compression portion 401.

[0072] Referring to Fig. 10B, the L* signal is supplied to a line buffer 401 in which it is transformed into blocks each of which is formed by 4 pixels x 4 pixels. The L* signal which has been sectioned to the above-described blocks is subjected to orthogonal transformation by the Hadamard converter 2 so as to be divided into DC component DC and AC component AC. The direct current component DC is transmitted to a scalar quantizer 403, while AC component is transmitted to a high frequency masking circuit 405, and a classifying device 404. The scalar quantizer 403 takes only MSB of 8 bits of the direct current component DC thus input so as to transmit it as a DC code.

[0073] The classifying device 404 classifies the input block data formed by 16 pixels into the following four classes in accordance with energy of the AC spectrum and the deviation of the same:
   Class 0...plain
   Class 1...vertical edge
   Class 2...lateral edge
   Class 3...diagonal

[0074] The above-described classes are expressed by using 2 bits.

[0075] The high frequency masking portion 405 receives the AC component from the Hadamard converter 2 and the class code expressed by 2 bits and transmitted from the classifying device 404 so that high frequency masking as shown by any of the hatching sections shown in Figs. 11A to 11D is performed in accordance with the value of the thus received code, namely the characteristics of image. Referring to Figs. 11A to 11D, the portions shown with mark x represent DC components, while the portions in which numerals are entered represent portions which are subjected to scalar quantization by the scalar quantizer 406.

[0076] The scalar quantizer 406 disposed next to the high frequency masking portion 405 also receive the above-described class codes each of which is formed by 2 bits and performs the scalar quantization of the bit distribution in any of manners shown in Figs. 11A to 11D in accordance with the received class codes, namely the characteristics of image. For example, if the received class is "1", the bit distribution is quantized in such a manner that the bits are arranged to be "4", "4", "3", "2" and "2" corresponding to five components A₁ to A₅. As a result, the total number of bits becomes 15.

[0077] A vector quantizing portion 407 is formed by a ROM 408 as shown in Fig. 12. The ROM 408 receives 15-bit data after the scalar quantization and class codes expressed by 2 bits in the form of a vector of 15 bits and outputs the index of 10 bits of the nearest vector in the table.

[0078] Data to be stored in the ROM 408 is the index of the nearest vector of the above-described vector supplied to the address portion of the ROM 408. That is, each of the states which can be supplied is expressed by a vector having 15-dimensional data (high frequency data included), and the vector which is the nearest, in terms of the spatial distance, to the input vector is previously calculated so as to cause the index of the vector which is the solution of the calculation to be stored. Since the high frequency band of the input data has been masked, only the low frequency band data is subjected to the calculation carried out prior to the retrieval for the nearest vector. In this case, each of the classes has its own code book, and each of the code books has 15-dimensional data.

[0079] When the sequency components A₁ to A₁₅ are defined and distributed as shown in Fig. 2A and 2B, the output vectors (A₁ -A₁₅) expressed by the 10-bit indexes in each of the classes "0" to "3", are as shown in Figs. 13A to 13D, respectively.

[0080] Thus, 8-bit DC codes are transmitted from the DC scalar quantizer 403, 2-bit class codes are transmitted from the classifying device 404, and 10-bit indexes are transmitted from the vector quantizer 407. Therefore, a total of 20 bits are, as the compressed data, transmitted from the L* compressor 401 so as to be written in the image memory 301 shown in Fig. 7.

[0081] Similarly to the L* compressor, the average value within the 4-by-4 pixel block, and the compressed code which has been obtained as the result of the above vector quantization are transmitted from the a*b* compressor 402. However, 12 bits are transmitted from the compressor 402. That is, codes of a total of 32 bits are transmitted from the compressor shown in Fig. 7 so as to be written in the memory 301.

[0082] Referring to Fig. 14, the expanding portion 304 shown in Fig. 7 will be described.

[0083] In the expanding portion 304, the compressed code is taken from the image memory 301 so as to be decoded. The 32-bit compressed code corresponding to the block of 4 x 4 pixels is divided into 20-bit L* compressed code and 12-bit a* and b* compressed code. The 20-bit L* compressed code is divided into 2-bit class code, 10-bit index code and 8-bit DC code.

[0084] The 8-bit DC code is decoded by being shifted to the left by 2 bits in a scalar inverse quantizer 410 shown in Fig. 14.

[0085] An AC decoder 409 is formed by a ROM and sectioned into four segments each of which is specified by a class code (2 bits), namely the characteristics of image. The 10-bit index code specifies one output vector (15 dimensional) in a segment specified by the class code. Each segment is composed similarly to in the decoder of the first embodiment shown in Fig. 5. An output vector from the segment is composed in a data structure which includes a high frequency component. The high frequency component to be added varies in accordance with the segment, namely the characteristics of image, even for the same index codes. Thus, the 15-dimensional AC data and the DC data which has been decoded by the inverse quantizer 410 are synthesized so as to generate a 4 x 4 complete matrix. The matrix data is supplied to a Hadamard inverse converter 411 so that a 4 x 4 actual image data block is transmitted.

[0086] Then, the classification performed by the classifying device 404 will be described.

[0087] Hadamard transformation according to this embodiment is an operation of orthogonal transformation performed on the basis of φ₀ to φ₁₅ shown in Fig. 15. That is, the 16-dimensional vector V which has been supplied is expressed by:


The input vector comprises all of the pixel values in a block of 4 pixels x 4 pixels. The pixel value is arranged to be 0 to 255 when expressed by 8 bits.

[0088] In this state, when a_i is obtained in accordance with:


a₀ can be expressed by 10 bits from 0 to 1023, while a₁ to a₁₅ can be expressed by 10 bits from -512 to 512.

[0089] In this state, L_s, V_s and H_s are respectively defined as follows as a physical quantity for classification:

[0090] Symbol L_s in Equation (14) shows the level of energy in the low frequency band which becomes enlarged when an sharp edge structure is included in the original image, V_s shows the intensity of the vertical line components in the original image and H_s shows the intensity of the horizontal line components in the original image. The characteristics of an input image are classified into the following four classes by using the above-described physical quantities:
   When Ls < 50, class 0: plain
   When LS ≧ 50 and also Vs > 2*Hs, class 1: vertical edge
   When LS ≧ 50 and also Hs < 2*Vs, class 2: horizontal edge
   When Ls > 50 and not classified into classes 1 and 2,
   class 3: diagonal edge

[0091] As a result of the above-described classification, the masking of the high frequency band can be performed in accordance with the energy distribution.

[0092] Meanwhile, although Hadamard transform is utilized as orthogonal transform in the above embodiments, a discrete cosine transform or discrete Fourier transform can also be utilized.

[0093] The DC scalar quantization is not limited to a linear quantization and non-linear quantization can be applied to the present invention. The quantizer can be comprised of ROM or RAM and so on.

[0094] The classification of an edge according to the present invention is not limited to plain, vertical, horizontal and diagonal. More specifically, the directions of the eage can be more finely classified. Also the class can be sorted into a text image, screened image and half tone image. Furthermore, the form of high frequency mask and the bit-assignment of AC components in the mask can vary in accordance with the class.

[0095] Furthermore, the present invention can also adopt a color ink-jet printer, color dot printer, color thermal transfer printer or color thermal printer as well as color laser printer. The present invention is especially effective in frame-sequenced recording method in which Y, M, C, K are sequentially recorded, especially in a method wherein the storage of one-frame image data is necessary.

[0096] The present invention is also not limited to the above embodiments and and therefore various changes and modifications can be made within the scope of the present invention as defined in the appended claims.

Claims

1. A decoding apparatus comprising means (9) for decoding a code which encodes a first set of frequency components (A₁, A₂, A₄, A₅, A₈) of a signal, characterised in that said decoding means (9) adds a second set of frequency components (A₃, A₄, A₇, A₉ -A₁₅) which was not encoded in said received code to the first set of frequency components using an assumed correlation between the first and second sets of frequency components.

2. The apparatus according to claim 1, wherein said first set of frequency components comprise orthogonally transformed blocks of signal data.

3. The apparatus according to claim 2, wherein said first set of frequency components further comprise a masked higher frequency component of a frequency component of the orthogonal transform.

4. The apparatus according to claim 2 or 3, wherein said correlation of said first and second sets of frequency components is between low and high frequency components within the signal.

5. The apparatus of any preceding claim wherein said signal is an image signal and said first and second sets of frequency components comprise spatial frequency components of a block of image data.

6. The apparatus according to claim 5, wherein said correlation of said first and second spatial frequency components is varied dependent on characteristics of the image data.

7. The apparatus according to claim 6, wherein said characteristics of the image data relates to the presence of an edge or the direction of an edge in the image data.

8. The apparatus according to any of claims 1 to 7, further comprising a memory (5b) for storing said correlation.

9. The apparatus according to any of claims 1 to 8, further comprising post-processing means (411) for processing an output of said decoding means.

10. The apparatus according to claim 9, wherein said post-processing means (411) performs an inverse orthogonal transform.

11. The apparatus according to any of claims 1 to 10, wherein said signal is an image signal, further comprising image forming means (200) for forming an image in accordance with an output of said decoding means (304).

12. The apparatus according to claim 11, wherein said image forming means (200) performs electro-photographic processing.

13. The apparatus according to any of claims 1 to 12, further comprising providing means (50/303) for providing said code via a transmission channel (23/301).

14. The apparatus according to claim 13, wherein said providing means includes an encoder (4, 303) for generating said code.

15. The apparatus according to claim 13, wherein said signal is an image signal and said providing means includes reading means (100) for reading an original to generate image data of said original.

16. The apparatus according to any of claims 1 to 15, wherein said code is encoded according to a vector-quantization method.

17. The apparatus according to any of claims 1 to 15, wherein said code represents image data comprising of a plurality of color components.

18. The apparatus according to claim 17, wherein said color components comprise brightness signal L*, color signal a* and color signal b* components.

19. A decoding method for recovering a signal from a received code, wherein said received code encodes a first set of frequency components of the signal, characterised in that a second set of frequency components which was not encoded in the received code is added to the first set of frequency components in accordance with an assumed correlation between the first and second sets of frequency components, and the first and second sets of frequency components are output together.

20. A decoding method according to claim 19, wherein said signal is an image signal and wherein said frequency components comprise spatial frequency components of a block of image data.

Ansprüche

1. Dekodiergerät mit einer Einrichtung (9) zum Dekodieren eines Codes, der einen ersten Satz von Frequenzkomponenten (A₁, A₂, A₄, A₅, A₈) eines Signals codiert,
dadurch gekennzeichnet, daß
die Dekodierereinrichtung (9) einen zweiten Satz von Frequenzkomponenten (A₃, A₆, A₇, A₉ bis A₁₅), der nicht in dem empfangenen Code codiert war, unter Verwendung einer angenommenen Korrelation zwischen dem ersten und zweiten Satz von Frequenzkomponenten zu dem ersten Satz von Frequenzkomponenten hinzufügt.

2. Gerät gemäß Patentanspruch 1, wobei der erste Satz von Frequenzkomponenten orthogonal transformierte Blöcke von Signaldaten umfaßt.

3. Gerät gemäß Patentanspruch 2, wobei der erste Satz von Frequenzkomponenten zudem eine maskierte höhere Frequenzkomponente einer Frequenzkomponente der orthogonalen Transformation umfaßt.

4. Gerät gemäß Patentanspruch 2 oder 3, wobei die Korrelation des ersten und zweiten Satzes von Frequenzkomponenten zwischen nieder- und hochfrequenten Komponenten innerhalb des Signals besteht.

5. Gerät gemäß einem der vorhergehenden Patentansprüche, wobei das Signal ein Bildsignal ist und der erste und zweite Satz von Frequenzkomponenten räumliche Frequenzkomponenten eines Bilddatenblocks umfaßt.

6. Gerät gemäß Patentanspruch 5, wobei die Korrelation der ersten und zweiten räumlichen Frequenzkomponenten abhängig von Charakteristiken der Bilddaten variiert wird.

7. Gerät gemäß Patentanspruch 6, wobei die Charakteristiken der Bilddaten sich auf das Vorhandensein einer Kante oder die Richtung einer Kante in den Bilddaten beziehen.

8. Gerät gemäß einem der Patentansprüche 1 bis 7, das zudem einen Speicher (5b) zum Speichern der Korrelation umfaßt.

9. Gerät gemäß einem der Patentansprüche 1 bis 8, das zudem eine Nachbearbeitungseinrichtung (411) zum Verarbeiten einer Ausgabe der Dekodierereinrichtung umfaßt.

10. Gerät gemäß Patentanspruch 9, wobei die Nachbearbeitungseinrichtung (411) eine inverse orthogonale Transformation durchführt.

11. Gerät gemäß einem der Patentansprüche 1 bis 10, wobei das Signal ein Bildsignal ist, das zudem eine Bilderzeugungseinrichtung (200) zum Ausbilden eines Bildes in Übereinstimmung mit einer Ausgabe der Dekodierereinrichtung (304) umfaßt.

12. Gerät gemäß Patentanspruch 11, wobei die Bilderzeugungseinrichtung (200) eine elektrofotografische Verarbeitung durchführt.

13. Gerät gemäß einem der Patentansprüche 1 bis 12, das zudem eine Bereitstellungseinrichtung (50/303) zum Bereitstellen des Codes über einen Übertragungskanal (23/301) umfaßt.

14. Gerät gemäß Patentanspruch 13, wobei die Bereitstellungseinrichtung einen Codierer (4, 303) zum Erzeugen des Codes enthält.

15. Gerät gemäß Patentanspruch 13, wobei das Signal ein Bildsignal ist und die Bereitstellungseinrichtung eine Leseeinrichtung (100) zum Lesen eines Originals enthält, um Bilddaten des Originals zu erzeugen.

16. Gerät gemäß einem der Patentansprüche 1 bis 15, wobei der Code gemäß einem Vektorquantisierungsverfahren codiert ist.

17. Gerät gemäß einem der Patentansprüche 1 bis 15, wobei der Code Bilddaten wiedergibt, die eine Vielzahl von Farbkomponenten umfassen.

18. Gerät gemäß Patentanspruch 17, wobei die Farbkomponenten ein Helligkeitssignal-L*-, Farbsignal-a*- und Farbsignal-b*-Komponenten umfassen.

19. Dekodierungsverfahren zum Rückgewinnen eines Signals von einem empfangenen Code, wobei der empfangene Code einen ersten Satz von Frequenzkomponenten des Signals codiert,
dadurch gekennzeichnet, daß
ein zweiter Satz von Frequenzkomponenten, der nicht in dem empfangenen Code codiert ist, in Übereinstimmung mit einer angenommenen Korrelation zwischen dem ersten und zweiten Satz von Frequenzkomponenten zu dem ersten Satz von Frequenzkomponenten hinzugefügt wird, und der erste und zweite Satz von Frequenzkomponenten zusammen ausgegeben wird.

20. Dekodierungsverfahren gemäß Patentanspruch 19, wobei das Signal ein Bildsignal ist und wobei die Frequenzkomponenten räumliche Frequenzkomponenten eines Bilddatenblockes umfassen.

Revendications

1. Appareil de décodage comprenant un moyen (9) pour décoder un code qui code un premier ensemble de composantes de fréquences (A₁, A₂, A₄, A₅, A₈) d'un signal, caractérisé en ce que ledit moyen (9) de décodage ajoute un second ensemble de composantes de fréquences (A₃, A₆, A₇, A₉, -A₁₅) qui n'ont pas été codées dans ledit code reçu, au premier ensemble de composantes de fréquences, par utilisation d'une corrélation supposée entre les premier et second ensembles de composantes de fréquences.

2. Appareil selon la revendication 1, dans lequel ledit premier ensemble de composantes de fréquences comprend des blocs de données de signaux soumis à une transformation orthogonale.

3. Appareil selon la revendication 2, dans lequel ledit premier ensemble de composantes de fréquences comprend en outre une composante de fréquence supérieure masquée d'une composante de fréquence de la transformée orthogonale.

4. Appareil selon la revendication 2 ou 3, dans lequel ladite corrélation desdits premier et second ensembles de composantes de fréquences s'effectue entre des composantes de fréquences faibles et élevées au sein du signal.

5. Appareil selon l'une quelconque des revendications précédentes, dans lequel ledit signal est un signal d'image et lesdits premier et second ensembles de composantes de fréquences comprennent des composantes de fréquences spatiales d'un bloc de données d'images.

6. Appareil selon la revendication 5, dans lequel ladite corrélation desdites première et seconde composantes de fréquences spatiales est amenée à varier en fonction de caractéristiques des données d'images.

7. Appareil selon la revendication 6, dans lequel lesdites caractéristiques de données d'images concernent la présence d'un bord ou la direction d'un bord dans les données d'images.

8. Appareil selon l'une quelconque des revendications 1 à 7, comprenant en outre une mémoire (5b) pour stocker ladite corrélation.

9. Appareil selon l'une quelconque des revendications 1 à 8, comprenant en outre un moyen (411) de post-traitement pour traiter une sortie dudit moyen de décodage.

10. Appareil selon la revendication 9, dans lequel ledit moyen (411) de post-traitement effectue une transformée orthogonale inverse.

11. Appareil selon l'une quelconque des revendications 1 à 10, dans lequel ledit signal est un signal d'image comprenant en outre un moyen (200) de formation d'image pour former une image conforme à une sortie dudit moyen (304) de décodage.

12. Appareil selon la revendication 11, dans lequel ledit moyen (200) de formation d'image effectue un traitement électro-photographique.

13. Appareil selon l'une quelconque des revendications 1 à 12, comprenant en outre un moyen (50/303) de fourniture pour fournir ledit code par l'intermédiaire d'un canal (23/301) de transmission.

14. Appareil selon la revendication 13, dans lequel ledit moyen de fourniture comporte un codeur (4, 303) pour générer ledit code.

15. Appareil selon la revendication 13, dans lequel ledit signal est un signal d'image et ledit moyen de fourniture comporte un moyen (100) de lecture pour lire un original afin de générer des données d'images dudit original.

16. Appareil selon l'une quelconque des revendications 1 à 15, dans lequel ledit code est codé conformément à un procédé de quantification vectorielle.

17. Appareil selon l'une quelconque des revendications 1 à 15, dans lequel ledit code représente des données d'images comprenant un ensemble de composantes de couleurs.

18. Appareil selon la revendication 17, dans lequel lesdites composantes de couleurs comprennent des composantes constituées par un signal de brillance L*, un signal de couleur a* et un signal de couleur b*.

19. Procédé de décodage pour reconstituer un signal à partir d'un code reçu, dans lequel ledit code reçu code un premier ensemble de composantes de fréquences du signal, caractérisé en ce qu'un second ensemble de composantes de fréquences n'ayant pas été codé dans le code reçu, est ajouté au premier ensemble de composantes de fréquences en conformité avec une corrélation supposée entre les premier et second ensembles de composantes de fréquences, et les premier et second ensembles de composantes de fréquences sont fournis ensemble en sortie.

20. Procédé de décodage selon la revendication 19, dans lequel ledit signal est un signal d'image et dans lequel lesdites composantes de fréquences comprennent des composantes de fréquences spatiales d'un bloc de données d'images.

Drawing